exploration

Files (1)

• SKILL.md

SKILL.md (first lines)

# Exploration

Produce a Candidate as a readable mathematical route: subproblem, origin, variables/objective/constraints, derivation, pseudocode, sources, assumptions, limitations, and a structural difference from known routes. An algorithm name, prompt persona, or parameter setting is not a distinct Candidate.

First-round authors receive only the problem contract, selected data/material, and reliable shared sources. Do not inject another first-round draft or recommendation. A human suggestion remains one candidate or a challenge target unless the human explicitly fixed a base route and its conditions independently hold.

Use a Variant only for implementation/configuration changes within a Candidate. A changed mathematical structure, stage composition, objective, or error propagation is a new Candidate; a fusion records parents but begins unreviewed.

Before formal scoring, fix data/split, metric formulas and units, constraints, meaningful baseline, fair budget, failure denominator, and stopping condition in a ComparisonContract. Preserve evidence, uncertainty, and tradeoffs in comparison packs. Do not manufacture aggregate scores, benchmarks, totals, or experimental conclusions.

From SoFarSoGoodya/mathmodel-weave, MIT.

More from SoFarSoGoodya/mathmodel-weave